Critical infrastructure faces mounting threats from supply chain attacks and AI-assisted intrusions that often go undetected due to visibility gaps in operational technology environments. Organizations struggle to detect and investigate these attacks because legacy systems lack proper monitoring, making it nearly impossible to determine the root cause or scope of compromises.
- Supply chain attacks like the Axios malware insertion affect thousands of systems before detection, with poisoned dependencies embedded in widely-used libraries.
- AI-assisted reconnaissance accelerates threat actor discovery of critical infrastructure from days of manual work to hours of automated analysis, as demonstrated in a Mexican water utility attack.
- Visibility gaps exist across two domains: operational technology (OT) environments with legacy equipment and the software supply chains that support them.
- Only 41% of organizations have built comprehensive AI security literacy, despite rapid agentic AI adoption in industrial landscapes.
- Key defenses include maintaining software bills of materials (SBOMs), implementing network segmentation, establishing baseline detection for anomalies, and maintaining transparent audit logs independent of vendor systems.
Critical infrastructure faces compounding threats: supply chain attacks like the Axios incident infiltrate thousands of systems before detection, while direct intrusions aided by AI, such as the attack on a Mexican municipal water utility, blend into complex operational technology (OT) environments. Many go undetected for extended periods, leaving organizations unable to determine what actually happened.
The core challenge is visibility. Modern OT systems consist of many interconnected components, including legacy equipment and systems that lack network and host monitoring. This makes it difficult to trace compromises back to their source. AI further compounds the problem, both as a tool attackers now exploit to accelerate reconnaissance, and as a capability that organizations deploy without adequate observability controls.
Evidence of visibility gaps
Recent real-world cases reveal a troubling pattern: despite significant investigation and response efforts, the root cause of disruption remains uncertain in many situations due to limited network and host visibility. These gaps manifest in two distinct ways: within OT environments themselves, and in the software supply chains upon which those environments increasingly depend.
Supply Chain compromise
In March, attackers published malicious versions of Axios, a widely used JavaScript library downloaded over 100 million times per week, by compromising a maintainer account. The poisoned releases inserted a hidden dependency that was executed during installation, enabling credential theft and remote access across affected systems.
This illustrates how deeply embedded third-party dependencies can be exploited in ways that are nearly impossible to detect downstream, especially when those dependencies flow into critical infrastructure environments.
The compromise was identified by security researchers who detected anomalous package behavior and unauthorized changes outside the normal development pipeline, prompting rapid removal. This highlights a structural visibility gap: even trusted, ubiquitous components can be altered in ways that evade detection and bypass standard development workflows, making it difficult for organizations to determine whether they were exposed or how far the compromise has spread.
AI-assisted infrastructure discovery
The visibility gap becomes more urgent when coupled with adversarial AI capabilities. In a recent case involving a Mexican municipal water utility, adversaries already inside the IT environment used AI to map the network and discover an industrial gateway, a SCADA/IIoT management interface connecting OT systems to IT. Without prior Industrial Control Systems (ICS) knowledge, the AI correctly identified it as a high-value target, analyzed its single-password authentication, and generated attack strategies.
Though the subsequent attack was unsuccessful, the incident demonstrates the core risk: AI accelerates information gathering from days of manual work to hours of automated analysis. For organizations with limited OT monitoring, this could mean the difference between early detection and unnoticed persistence.
Long-running persistence
Advanced attackers can keep quiet, persistent to critical environments over extended periods. Remaining unnoticed for a long time and blending into normal business activity, detection and full removal become more challenging. When combined with the OT complexity discussed below, such persistence becomes nearly impossible to fully characterize.
The rise of complex incidents
The 2017 TRITON attack on industrial safety systems highlighted how OT network and host visibility gaps can obscure what happened, leaving scope, intent, and impact unclear. Why is this the case? The OT used in critical infrastructure is becoming increasingly complex, requiring people, processes, and technology to work in unison to detect adversaries. As organizations rely on interconnected systems spanning multiple vendors, teams, and generations of technology, full end-to-end oversight is increasingly difficult.
OT automation and control systems increasingly rely on complex software supply chains to expand capabilities and speed up development. But that integration brings new security risks. As software becomes more embedded into critical infrastructure operations, organizations face two compounding visibility challenges: maintaining visibility into their operational systems during incidents, and into the software supply chains that support them. Gaps in either area make it difficult to determine the root cause and scope of a disruption. This is why a comprehensive security framework matters, covering monitoring, detection, and evaluation.
AI is accelerating the OT complexity problem
Organizations are now adopting agentic AI at a rapid rate to accelerate efficiencies across the industrial landscape. However, SANS data shows that less than half of organizations (41%) have built any form of comprehensive AI security literacy. This creates a mounting risk: while AI is being implemented in cybersecurity and business workflows, there is insufficient training and education to ensure visibility and control.
The risk manifests in two ways. First, adversaries can use AI systems to conduct more sophisticated reconnaissance than ever before. Second, AI-driven automation can increase efficiency for defenders, but reduce the structural visibility of system behavior during failure events. Organizations implementing agentic AI without adequate logging and transparency, particularly when relying on third-party vendors whose systems operate as "black boxes," further exacerbate the problem. The speed of adoption also increases risk from vendor instability: a market correction could leave organizations dependent on AI-augmented automation from vendors that may not survive the downturn. If those systems become unsupported, the observability and audit data needed to investigate disruptions may no longer be accessible, leaving incidents with no explanation.
Defending against system failures and cyberattacks
The true cost of attacks is indirect business impacts, ranging from lost production and broken supply commitments to erosion of customer trust, all of which far exceed the cost of the attack itself. Organizations must implement layered defenses addressing the specific visibility gaps identified above:
For supply chain risks: Maintain an SBOM and component-integrity tracking. Test updates in isolated environments before production. Establish baseline detection for unexpected dependencies or build-time changes.
For AI-assisted reconnaissance: Assume threat actors will use AI for OT discovery. Detect reconnaissance patterns, unusual network scans, gateway probing, and authentication attempts, rather than relying solely on prevention. Implement network segmentation to make OT assets harder to enumerate.
For agentic AI you deploy: Maintain transparent audit logs independent of vendor systems. Implement least-privilege access and formal governance to ensure AI decisions remain traceable and investigable.
Organizations must embed these controls from the start, not retrofit them later. These are not compliance checkboxes; they are what separates an incident you can learn from vs. one you can never fully explain.
Conclusion
The rise of complex incidents signals a shift in critical infrastructure risk, underscoring the fundamental ways attack techniques are evolving and how organizations must keep up. It is paramount for organizations to become aware of the risks posed by both OT complexity and AI adoption, and to implement frameworks that maintain visibility across both domains.
While AI integration is inevitable, prioritizing visibility across the lifecycle of OT systems is equally crucial for ensuring the efficiency and profitability of OT environments. Organizations must ensure that when disruptions do occur, they have visibility in place to quickly identify the root cause, learn from it, and incorporate lessons to prevent it from happening again. The cost of visibility is far less than the cost of an unexplained incident.
